Leveraging Two-Stage Transfer Learning with YOLOv7 and CNN-Based LLI Enhancement for Trike Detection

The initial accessibility of electric tricycles, also known as “e-trikes,” had led to a pattern of many untrained drivers on the road, which made e-trikes a subject of scrutiny in the Philippines. Consequently, the Philippine government responded to the wave of e-trike-related accidents with a regulation prohibiting e-trikes as well as e-bikes, tricycles, and other similar vehicles on national roads, circumferential roads, and radial roads in Metro Manila. To help aid this solution, this work utilized a low-light image enhancement solution and a two-stage transfer learning method to create a YOLOv7 model that detects e-trikes and tricycles with 98.2% precision and 96.2% recall overall, as well as a 98% and a 99% accuracy for each class, respectively.